Comparing the Twitter Response Times of the Public and Local Officials During the Camp Fire

Olivia Plazanet, Madeline Hoshko and Taylor Hinchliff

Introduction and Background

Introduction

  • We will be examining the response times of several “official” Twitter accounts: CalFire, Butte County Sheriff, and Chico Fire Department.
  • Why is this important?
    • Natural disaster require timely warnings and information. This makes it easier to overcome public misperception and helps different actors in responding to crises in a way that benefits the collective.
  • How credible is this information?
  • What about using television? Phone calls? Radio?
  • That is why it is also very important to examine the times it takes for officials to issue official information and warnings to the public, such as evacuation notices, which can literally be life-saving information during disasters.

Why Use Twitter?

  • Fewer people are using television and radio than ever before.
  • Over reliance on phone calls or texts for emergency information (St. John, Paige, and Joseph Serna, 2018)
  • Twitter has 64.2 million users, and a diverse user base.
  • 77% of journalists say social media is important to them for learning about potential stories more quickly, and that it is important in reporting stories more quickly (Elliott 2018).
  • During November of 2018 alone, CalFire’s official Twitter account gained about 35k followers, which shows them to be an important source of information for the public (SocialBlade, 2020).

Background and Lit Review

  • Disaster Response and Twitter
    • Why is Twitter data important during disasters?
      • Gives a “real-time view” of disasters, allows fast-flowing information quickly from official sources (Ford, 2018).
    • There is a need for unique updates across organizations given the volume of data produced because manually managing is not always practical right after a disaster (Ford, 2018).
  • Potential Flaws in Phone-Based Warning Systems
    • “Butte County Sheriff’s office showed that the first evacuation orders requested by firefighters failed to connect. Half of the calls went to voicemail (St. John and Serna, 2018).”
    • This inefficiency presents a real problem within our system.
    • The police chief of Paradise said “there was no time for a citywide evacuation order — the city’s own system went down in the midst of a partial order” (St. John and Serna, 2018). With California having the most Twitter users, there needs to be a wide array of emergency notification methods through Twitter (Stirtz, 2020).

Background and Lit Review

  • Refiguration of disaster social relations
    • New communication technologies replacing traditional media, transformation of disaster visibility (Cottle 2014).
    • Individuals are central in emergency responses (Cottle 2014).
    • Civilian surge of information and worldwide engagement because of fast-flowing information, collaboration between citizens, disaster relief agencies, volunteers, and public figures (Cottle 2014).
  • Limits to crisis data
    • When data collection starts, social media platforms can create analytical and ethical oversights (limited access to the internet during a disaster: power outages, etc) (Crawford and Finn 2015).
    • Traditional media was state-controlled, but that changed in more recent years (Crawford and Finn 2015).
    • Traditional media was state-controlled, but that changed in more recent years (Crawford and Finn 2015).
    • People cannot control how their data is used: privacy issue (Crawford and Finn 2015).

Research Problem and Topic Area

  • How timely were public officials and emergency services twitter accounts in disseminating important and crucial information to the public during the Camp Fire?
  • Most research done revolves around incidents such as earthquakes (only a sixth of cases pertained to wildfires).
  • Our main goal is to examine the timeliness of social media in emergency management by examining the response times of specific accounts along the timeline of the Camp Fire.
  • We will use a timeline to examine these response times more closely and compare them to the timeline of the actual disaster.

Data Introduction

  • 72,748 tweets were collected using the rtweet (Kearney MW 2019) package using R 3.6.2 (R Core Team) beginning the morning of November 11, 2018 by the Spring 2020 Advanced Data Science class at CSU, Chico. Tweets were collected Using the query: #CampFire, #campfire, #paradise, #Paradise or @CALFIRE_ButteCo"
  • A timeline of events from the morning of November 8, 2018 was derived from the Chico Enterprise-Record. (Epley)
  • We are observing tweets regarding seven public figures’ and organizations’ twitter accounts that were instrumental during the Camp Fire, specifically: @Cal_Fire, @CalFire_ButteCo, @ButteSherriff, @ChicoFD, @CountyOfButte, @Paradise_CA, and @ChicoPolice.
    • Data from these accounts are defined as “media” while all others are defined as “public”.

Methods

  • Variable Creation: To successfully compare the response time of the local media accounts to the real life timing of events we populated a data frame with each event and time as specified by Chico Enterprise-Record’s timeline of the morning of the fire (Epley).
  • Data Analysis: We analyzed the specific timing of emergency services and government entities’ tweets regarding pertinent information in relationship to the real life events through a timeline that incorporates both the twitter data and the derived timeline data in one merged timeline plot.
    • We additionally created a timeline documenting the comparison between times of general public tweets and tweets from our local media sources.

Description of variables being used

  • We were most interested in initial response times data, so we filtered down the dataset to only include tweets that occurred during the first six hours from the fire starting.
  • Each record in the data represents an individual tweet’s metadata. The variables from the metadata used throughout include:
    • Tweet content
    • User’s Twitter Handle
    • Whether the tweet was a retweet
    • Time the tweet was created
    • The events derived from Chico Enterprise-Record’s timeline (Epley)

Results

Univariate Description of Measures

While the original dataset includes 72,748 tweets, we are only concerned with the tweets that occur within the first 6 hours of the fire starting and those that directly address the fire. This ends up being just 446 tweets, which amounts to 0.61% of the overall data.

Explanatory Variable Description:

  • Our explanatory variable is the local media accounts. Of the 446 tweets 425 (95.29%) are classified as public and 21 (4.71%) are classified as media (tweets from our local media accounts).

Response Variable Description:

  • Our response measure is the time of the tweets.

  • The time of the first tweet was 11/08/2018 06:51:47

Bivariate Description of Response

The response times and amount of tweets for each account can be shown here to display the local media accounts that were most involved within the first hours of the fire.
The media’s first response to the fire on Twitter appears at 06:51:47 by CAL FIRE Butte County


User Time of First Tweet
ButteSheriff 2018-11-08 07:23:02
CALFIRE_ButteCo 2018-11-08 06:51:47
ChicoFD 2018-11-08 10:46:27

Public Tweets

Here we can see the activity of public accounts within the first six hours. All retweets are retweets of our local media accounts.

The first tweet by a public user closely followed CAL FIRE Butte County’s at 06:54:55

The Media Response Compared to Real-Time Events

In the following timeline each dot signifies a tweet or event to show tweet times in comparison to the real event times. The majority of the tweets shown are evacuation warnings and orders.

The Media Response Compared to Public Response

After eliminating retweets for public tweets the following timeline compares the timing of tweets by the public vs the local media for the morning up until 10:00 AM.

What Does it All Mean?

  • Our results show that Twitter can be used to disseminate important and potentially life saving information in a timely manner, as was seen on the timeline of events for the Camp Fire.
  • From our Twitter data, we can also see that Twitter can provide individuals with this information in a consistent manner, when other media methods may be ineffective in sharing this data (such as losing power during a disaster and being unable to receive news via television).
  • This ultimately means that social media platforms such as Twitter, if used properly, can be an important tool when it comes to emergency alert systems.

Why Should We Care?

  • When official accounts respond to events promptly, it gives the organization or person more credibility, especially in the context of natural disasters where people count on them to stay safe and informed on the changing situation.
  • With a mounting number of people using Twitter, data collected from social media is useful in offering insight into public opinion.
  • Retweets permeate information at a rapid pace, which enables a message to be communicated to a large number, of people, which can potentially limit disaster.

References

Ceron, Andrea. (2014). Twitter and the Traditional Media: Who is the Real Agenda Setter?. APSA 2014 Annual Meeting Paper

Cottle, Simon. (2014). Rethinking Media and Disasters in a Global Age: What’s Changed and Why It Matters. Media, War & Conflict. 7. 3-22.10.1177/1750635213513229. https://www.researchgate.net/publication/270633894_Rethinking_Media_and_Disasters_in_a_Global_Age_What's_Changed_and_Why_It_Matters

Crawford, Kate and Finn, Megan. (2015). The limits of crisis data: analytical and ethical challenges of using social and mobile data to understand disasters. GeoJournal. 80. 491-502 https://www.researchgate.net/publication/285401858_The_limits_of_crisis_data_analytical_and_ethical_challenges_of_using_social_and_mobile_data_to_understand_disasters

Elliott, Jennifer. “8 Best Practices for Emergency Communications on Social Media.” EfficientGov, 19 July 2018, https://www.efficientgov.com/community-engagement/articles/8-best-practices-for-emergency-communications-on-social-media-vwZS7OO5eoblrs3G/.

Epley, Robin (November 8, 2019). “Timeline: Breaking down Nov. 8 - the Day the Camp Fire Sparked.” Chico Enterprise-Record, Chico Enterprise-Record https://www.chicoer.com/2019/11/07/timeline-breaking-down-nov-8-the-day-the-camp-fire-sparked/.

Ford, Jordan. “Improving Disaster Response through Twitter Data.” Penn State University, 2018, https://news.psu.edu/story/527730/2018/07/10/research/improving-disaster-response-through-twitter-data.

References cont.

Kearney MW (2019). “rtweet: Collecting and analyzing Twitter data.” Journal of Open Source Software, 4(42), 1829. Doi: 10.21105/joss.01829, R package version 0.7.0, https://joss.theoj.org/papers/10.21105/joss.01829.

R Core Team (2019). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org.

Radio Ink (March 3, 2020). “The Decline of the Home Radio”. https://radioink.com/2020/03/03/the-decline-of-the-home-radio/

Rice, Doyle (January 8, 2019). “USA had world’s 3 costliest natural disasters in 2018, and Camp Fire was the worst”. USA Today.

Social Blade. “Cal Fire’s Twitter Stats Summary Profile.” Social Blade, 2020, socialblade.com/twitter/user/cal_fire.

St. John, Paige, and Joseph Serna. “Camp Fire Evacuation Warnings Failed to Reach More than a Third of Residents Meant to Receive Calls.” Los Angeles Times, 1 Dec. 2018, https://www.latimes.com/local/california/la-me-ln-paradise-evacuation-warnings-20181130-story.html.

Stirtz, Kevin. “Twitter Ranking: Which States Twitter the Most?” All Business, Dun & Bradstreet, 14 May 2009, https://www.allbusiness.com/twitter-ranking-which-states-twitter-the-most-12329567-1.html.